Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Invariances in Neural Networks from Training Data
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use 4 augmented copies instead of 1 for Augerino (ablation) | Augerino · CIFAR-10 test | accuracy: 92.22 -> 93.81 | Helped |
| learn augmentation distribution (Augerino, 4 copies) instead of fixed augmentation | 13-layer CNN · CIFAR-10 test | accuracy: 92.64 -> 93.81 | Helped |
| learned augmentation (4 copies) instead of Fast AutoAugment | 13-layer CNN · CIFAR-10 test | accuracy: 92.65 -> 93.81 | Helped |
| add Augerino augmentation vs no augmentation | LieConv-Trivial · QM9 HOMO | MAE (meV): 52.7 -> 38.3 | Helped |
| add Augerino augmentation vs no augmentation | LieConv-T(3) · QM9 HOMO | MAE (meV): 34.2 -> 33.2 | Helped |
| add Augerino augmentation vs no augmentation | LieConv-Trivial · QM9 LUMO | MAE (meV): 43.5 -> 33.7 | Helped |
| hand-chosen SE(3) augmentation instead of learned Augerino | LieConv-Trivial · QM9 HOMO | MAE (meV): 38.3 -> 36.5 | Helped |
| learn brightness and contrast invariance with Augerino | 13-layer CNN · STL-10 test | accuracy (%): 89.0 -> 89.7 | Helped |
| vary regularization weight lambda (0.01, 0.05, 0.1) (ablation) | Augerino · rotMNIST | accuracy | No clear effect |
| add invariance-promoting regularization to Augerino (ablation) | Augerino · CIFAR-10 test | accuracy | Helped |
Closest papers by meaning
- Exploiting Invariance in Training Deep Neural Networks (AAAI 2022 · similarity 0.92)
- Grounding inductive biases in natural images: invariance stems from variations in data (NeurIPS 2021 · similarity 0.91)
- In What Ways Are Deep Neural Networks Invariant and How Should We Measure This? (NeurIPS 2022 · similarity 0.91)
- Deep invariant networks with differentiable augmentation layers (NeurIPS 2022 · similarity 0.89)
- On Making $SE(2)$-Invariant Networks Optimal (NeurIPS 2026 · similarity 0.88)
- From Bricks to Bridges: Product of Invariances to Enhance Latent Space Communication (ICLR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.